Schedule of Class Meeng

Mul$media  Networking  

  #9  CDN  Solu$ons   Semester  Ganjil  2012   PTIIK  Universitas  Brawijaya  

  

Schedule  of  Class  Mee$ng  

1.

    Introduc$on     2.

    Applica$ons  of  MN   3.

    Requirements  of  MN   4.

    Coding  and   Compression  

  5.

    RTP   6.   IP  Mul$cast   7.

   

  IP  Mul$cast  (cont’d)     8.

    Overlay  Mul$cast   9.

    CDN:  Solu,ons  

  10. CDN:  Case  Studies  

  11. QoS  on  the  Internet:   Constraints  

  12. QoS  on  the  Internet:   Solu$ons  

  13. Discussion   14.

   Summary  

  

Today’s  Outline  

• Ways  to  distribute  video  online  
• –   IP  Mul$cast  

    Client-­‐server  

• –   P2P  Media  Streaming  
• – –  CDN  (Content  Delivery  Networks)  

  Content  Delivery/Distribu$on   Networks  (CDN)   Content  Distribu$on  Network   how  to  stream  content  (selected     challenge:  

• from  millions  of  videos)  to  hundreds  of   thousands  of  simultaneous  users?   single,  large  “mega-­‐server”  
• –  single  point  of  failure  

    op,on  1:  

• –  point  of  network  conges$on  
• –  long  path  to  distant  clients  

• –  mul$ple  copies  of  video  sent  over  outgoing  link  

  ….quite  simply:  this  solu$on   doesn’ t  scale  

  Content  Distribu$on  Network  

• challenge:   how  to  stream  content  (selected  from  

    millions  of  videos)  to  hundreds  of  thousands  of   simultaneous  users?  

• op,on  2:  

  store/serve  mul$ple  copies  of  videos  at     mul$ple  geographically  distributed  sites   (CDN)  

• – push  CDN  servers  deep  into  many  access  

    enter  deep:   networks    

• close  to  users  
• used  by  Akamai,  1700  loca$ons  
• – smaller  number  (10’s)  of  larger  clusters  in  

    bring  home:   POPs  near  (but  not  within)  access  networks  

    

• used  by  Limelight

Content  Distribu$on  Network  

  origin server in North America CDN distribution node CDN server in S. America CDN server in Europe CDN server in Asia

• Proac$ve  content  replica$on  
• –  Content  provider  (e.g.,  CNN)   contracts  with  a  CDN  
• CDN  replicates  the  content    
• –  On  many  servers  spread   throughout  the  Internet  
• Upda$ng  the  replicas  
• –  Updates  pushed  to  replicas   when  the  content  changes  

  Server  Selec$on  Policy  

• how  does  CDN  DNS  select  “good”  

    challenge:   CDN  node  to  stream  to  client  

• –  pick  CDN  node  geographically  closest  to  client  

• –  pick  CDN  node  with  shortest  delay  (or  min  #  hops)  

  to  client  (CDN  nodes  periodically  ping  access  ISPs,   repor$ng  results  to  CDN  DNS)  
• –  IP  anycast  

  client  decide  -­‐  give  client  a  list     alterna,ve:   of  several  CDN  servers  

• let  
• –  client  pings  servers,  picks  “best”  

    

• –  Nedlix  approach  

  

Server  Selec$on  Policy  

Requires continuous monitoring of

• Live  server  

  liveness, load, and performance

• –  For  availability  
• Lowest  load  
• –  To  balance  load  across  the  servers  
• Closest  
• –  Nearest  geographically,  or  in  round-­‐trip  $me  
• Best  performance  
• –  Throughput,  latency,  …  
• Cheapest  bandwidth,  electricity,  …  

  Server  Selec$on  Mechanism  

• Applica$on   •  Advantages  
• –  HTTP  redirec$on   –  Fine-­‐grain  control  
• –  Selec$on  based  on   client  IP  address  

GET

• Disadvantages  

Redirect

• –  Extra  round-­‐trips  for  

GET

  TCP  connec$on  to   server   OK

• –  Overhead  on  the  server  

Server  Selec$on  Mechanism  

• Rou$ng  
• Advantages  

• –  Anycast  rou$ng   –  No  extra  round  trips  

• –  Route  to  nearby  server  
• Disadvantages  

  1.2.3.0/24

• –  Does  not  consider   network  or  server  load  
• –  Different  packets  may   go  to  different  servers  
• –  Used  only  for  simple  

  1.2.3.0/24 request-­‐response  apps  

  

Server  Selec$on  Mechanism  

• Naming   •  Advantages  
• –  DNS-­‐based  server   –  Avoid  TCP  set-­‐up  delay   selec$on  
• –  DNS  caching  reduces   overhead  

  1.2.3.4

• –  Rela$vely  fine  control  
• Disadvantage  

DNS query

• –  Based  on  IP  address  of   local  DNS  server  
• –  “Hidden  load”  effect  

  1.2.3.5 local DNS server

• –  DNS  TTL  limits  adapta$on  

How  Akamai  Works  

  Akamai  Sta$s$cs  

• Distributed  servers   •  Client  requests  
• – Servers:  ~61,000   – Hundreds  of   billions  per  day  
• – Networks:  ~1,000 &n
• – Half  in  the  top    
• – Countries:  ~70   45  networks  
• Many  customers  
• – 15-­‐20%  of  all  Web  
• – Apple,  BBC,  FOX,  GM   traffic  worldwide  

  IBM,  MTV,  NASA,   NBC,  …   HTTP

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server

  1

  2 GET index.

  html http://cache.cnn.com/foo.jpg HTTP  

  Akamai cluster

  Akamai global DNS server Akamai regional DNS server

  Nearby Akamai cluster

End user

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server

  DNS lookup cache.cnn.com

  Akamai

  Akamai global

  cluster

  DNS server

  3

  1

  ALIAS:

  4 Akamai regional

  g.akamai.net DNS server

  Nearby Akamai cluster

  End user

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server

  DNS lookup g.akamai.net

  Akamai

  Akamai global

  cluster

  5 DNS server

  3

  1

  6

  4 Akamai regional

  ALIAS DNS server a73.g.akamai.net

  Nearby Akamai cluster

  End user

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server Akamai

  Akamai global

  cluster

  5 DNS server

  3

  1

  6

  4 Akamai regional

  7 DNS server

  net mai. g.aka

  8 a73.

  DNS

  Nearby

  Address

  Akamai

  1.2.3.4

  cluster

  End user

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server Akamai

  Akamai global

  cluster

  5 DNS server

  3

  1

  6

  4 Akamai regional

  7 DNS server

  8 Nearby Akamai

  9 cluster

  End user GET /foo.jpg Host: cache.cnn.com

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server

  GET foo.jpg

  11

  12 Akamai

  Akamai global

  cluster

  5 DNS server

  3

  1

  6

  4 Akamai regional

  7 DNS server

  8 Nearby Akamai

  9 cluster

  End user GET /foo.jpg Host: cache.cnn.com

  How  Akamai  Uses  DNS  

  cnn.com (content provider) DNS root server

  11

  12 Akamai

  Akamai global

  cluster

  5 DNS server

  3

  1

  6

  4 Akamai regional

  7 DNS server

  8 Nearby Akamai

  9 cluster

  End user

  10

  

How  Akamai  Works:  Cache  Hit  

  cnn.com (content provider) DNS root server Akamai server

  1

  2 Akamai high-level

  DNS server Akamai low-level DNS server

  Nearby hash-chosen Akamai server

  7

  8

  9

  10 GET index.

  html GET /cnn.com/foo.jpg

  End user

  

Mapping  System  

• Equivalence  classes  of  IP  addresses  

• –  IP  addresses  experiencing  similar  performance  

• –  Quan$fy  how  well  they  connect  to  each  other  
• Collect  and  combine  measurements  
• –  Ping,  traceroute,  BGP  routes,  server  logs  
• E.g.,  over  100  TB  of  logs  per  days  
• –  Network  latency,  loss,  and  connec$vity  

   

  

Mapping  System  

• •  Map  each  IP  class  to  a  preferred  server  cluster  

• –  Based  on  performance,  cluster  health,  etc.  
• –  Updated  roughly  every  minute      
• Map  client  request  to  a  server  in  the  cluster  
• –  Load  balancer  selects  a  specific  server  
• –  E.g.,  to  maximize  the  cache  hit  rate  

  

Adap$ng  to  Failures  

• Failing  hard  drive  on  a  server  

• –  Suspends  aqer  finishing  “in  progress”  requests  

• Failed  server  
• –  Another  server  takes  over  for  the  IP  address  
• –  Low-­‐level  map  updated  quickly  
• Failed  cluster  
• –  High-­‐level  map  updated  quickly  
• Failed  path  to  customer’s  origin  server  
• –  Route  packets  through  an  intermediate  node  

  

Akamai  Transport  Op$miza$ons  

• Bad  Internet  routes  
• –  Overlay  rou$ng  through  an  intermediate  server  
• Packet  loss  
• –  Sending  redundant  data  over  mul$ple  paths  
• TCP  connec$on  set-­‐up/teardown  
• –  Pools  of  persistent  connec$ons  
• TCP  conges$on  window  and  round-­‐trip  $me  

• –  Es$mates  based  on  network  latency  measurements  

     

  

Akamai  Applica$on  Op$miza$ons  

• Slow  download  of  embedded  objects  
• –  Prefetch  when  HTML  page  is  requested  
• Large  objects  
• –  Content  compression  
• Slow  applica$ons  
• –  Moving  applica$ons  to  edge  servers  
• –  E.g.,  content  aggrega$on  and  transforma$on  
• –  E.g.,  sta$c  databases  (e.g.,  product  catalogs)  

• –  E.g.  batching  and  valida$ng  input  on  Web  forms  

  

Conclusion  

• Content  distribu$on  is  hard  
• –  Many,  diverse,  changing  objects  
• –  Clients  distributed  all  over  the  world  
• –  Reducing  latency  is  king  
• Contribu$on  distribu$on  solu$ons  
• –  Reac$ve  caching  
• –  Proac$ve  content  distribu$on  networks  
CDN:  “simple”  content  access  scenario   hrp://netcinema.com

  Bob  (client)  requests  video   /6Y7B23V   video  stored  in  CDN  at  hrp://KingCDN.com/NetC6y&B

  23V  

     1. Bob gets URL for for video http://netcinema.com/6Y7B23V 2. resolve http://netcinema.com/6Y7B23V from netcinema.com via Bob’s local DNS web page

  2

  1

  5 6. request video from

  4&5. Resolve KINGCDN server, http://KingCDN.com/NetC6y&B23 streamed via HTTP via KingCDN’s authoritative DNS,

  3. netcinema’s DNS returns URL

  netcinema.com

  which returns IP address of KingCDN

  4 http://KingCDN.com/NetC6y&B

  

23V

  server with video

  3

  netcinema’s KingCDN

  Case  study:  Nedlix  

• 30%  downstream  US  traffic  in  2011  

  rd  party  

• owns  very  lirle  infrastructure,  uses  3 services:  
• –  own  registra$on,  payment  servers  

  rd  party)  cloud  services:  

• –  Amazon  (3
• Nedlix  uploads  studio  master  to  Amazon  cloud  
• create  mul$ple  version  of  movie  (different  endodings)  in   cloud  
• upload  versions  from  cloud  to  CDNs  
• Cloud  hosts  Nedlix  web  pages  for  user  browsing  

  rd  3  party  CDNs  host/stream  Nedlix  content:     three

• – Akamai,  Limelight,  Level-­‐3  

Case  study:  Nedlix  

  1

  1. Bob manages Netflix account

  Netflix registration, accounting servers Amazon cloud

  Akamai CDN Limelight CDN Level-3 CDN

  2

  2. Bob browses Netflix video

  3

  3. Manifest file returned for requested video

  4. DASH streaming upload copies of multiple versions of video to CDNs